Temporal Flexibility in Spiking Neural Networks: Towards Generalization Across Time Steps and Deployment Friendliness

Spiking Neural Networks (SNNs), models inspired by neural mechanisms in the brain, allow for energy-efficient implementation on neuromorphic hardware. However, SNNs trained with current direct training approaches are constrained to a specific time step. This "temporal inflexibility" 1) hinders SNNs' deployment on time-step-free fully event-driven chips and 2) prevents energy-performance balance based on dynamic inference time steps. In this study, we first explore the feasibility of training SNNs that generalize across different time steps. We then introduce Mixed Time-step Training (MTT), a novel method that improves the temporal flexibility of SNNs, making SNNs adaptive to diverse temporal structures. During each iteration of MTT, random time steps are assigned to different SNN stages, with spikes transmitted between stages via communication modules. After training, the weights are deployed and evaluated on both time-stepped and fully eventdriven platforms. Experimental results show that models trained by MTT gain remarkable temporal flexibility, friendliness for both event-driven and clock-driven deployment (nearly lossless on N-MNIST and 10.1% higher than standard methods on CIFAR10-DVS), enhanced network generalization, and near SOTA performance. To the best of our knowledge, this is the first work to report the results of large-scale SNN deployment on fully event-driven scenarios. As deep learning continues to evolve, the field has witnessed numerous groundbreaking advancements that have made unprecedented strides across diverse applications. However, deploying these huge neural networks on low-power edge devices presents substantial challenges. In addition to the typical solutions, including network quantization (Rastegari et al., 2016), pruning (He et al., 2017), and distillation (Hinton et al., 2015), the Spiking Neural Networks, known as one of the 3rd generation of neural networks, have emerged as a compelling candidate due to their unique bio-inspired characteristics (Fang et al., 2021; Guo et al., 2022; Yao et al., 2023). SNNs mimic the behavior of biological neurons by accumulating membrane potentials and transmitting sparse spikes, thereby circumventing the need for computationally expensive multiplications (Roy et al., 2019), which presents a promising solution for energy-efficient neuromorphic computation.